Announcement

Collapse
No announcement yet.

OpenMP 3.1 Support Readied By Intel For LLVM Clang

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • OpenMP 3.1 Support Readied By Intel For LLVM Clang

    Phoronix: OpenMP 3.1 Support Readied By Intel For LLVM Clang

    Intel software engineers have implemented full support for OpenMP 3.1 onto LLVM's Clang C/C++ compiler front-end...

    http://www.phoronix.com/vr.php?view=MTQ0NjQ

  • #2
    Now the question is: will the maintainers accept the patches upstream ? It is pretty well known that they want to scale to one million of threads

    Comment


    • #3
      Now we just need some LLVM Fortran love

      Comment


      • #4
        Here's hoping for merging in time for 3.4.

        Comment


        • #5
          Millions of threads and OpenMP

          Originally posted by wargames View Post
          It is pretty well known that they want to scale to one million of threads
          I've seen this multiple times here on Phoronix. Care to provide a reference to the "million of threads" remark? I read every email on the llvm/cfe mailing lists (dev and commits) and have never come across this sentiment/desire there. It is entirely possible I missed it.

          Tangentially, if you take a look at the current OpenMP thread, reviewer bandwidth is cited as the main obstacle keeping OpenMP support from finding its way upstream, not some philosophical/ideological objection.

          Thanks!

          Carlos

          Comment


          • #6
            Awesome that it was started by AMD and then completed by Intel. I can't really remember the last time they worked together on something like this!

            Comment


            • #7
              Originally posted by sandain View Post
              Now we just need some LLVM Fortran love
              It's a GSoC project.

              Alex Lorenz is working on it.

              It's called Flang.

              Hal Finkel is his mentor.

              https://www.google-melange.com/gsoc/.../gsoc2013/llvm

              Comment


              • #8
                Originally posted by cjmacedo View Post
                I've seen this multiple times here on Phoronix. Care to provide a reference to the "million of threads" remark? I read every email on the llvm/cfe mailing lists (dev and commits) and have never come across this sentiment/desire there. It is entirely possible I missed it.

                Tangentially, if you take a look at the current OpenMP thread, reviewer bandwidth is cited as the main obstacle keeping OpenMP support from finding its way upstream, not some philosophical/ideological objection.

                Thanks!

                Carlos
                http://phoronix.com/forums/showthrea...807#post326807

                NOTE: A "million" is obviously an exageration...

                Comment


                • #9
                  Originally posted by wargames View Post
                  http://phoronix.com/forums/showthrea...807#post326807

                  NOTE: A "million" is obviously an exageration...
                  Thank you, I finally got to the root of this misunderstanding that plagues this forum (and even worse, has always plagued me! )

                  So further along in that same thread this was offered as "proof"

                  It happens to be the case that this is NOT an LLVM document. This site hosts the ideas of Preston Briggs, an LLVM contributor that AFAICT never replied on any of the OpenMP threads on the lists. His contributions have been limited to Dependency Analysis.

                  From comments I've read elsewhere (which I rather not link but are readily available to those who know where to look) some LLVM devs do in fact believe that OpenMP probably isn't the best solution to the problems it attempts to solve. Thread count scalability isn't cited as a reason.

                  I highly doubt OpenMP won't make it into the main repo. Not only has upstreaming already begun but even more tellingly, it seems to have happened with the help (i.e. reviews) of some core developers that initially pushed-back on its inclusion in Clang.

                  Cheers,

                  Carlos

                  Comment


                  • #10
                    Finally!!!! :d

                    I swear this is probably the BIGGEST and most important advancement in linux this year for me. This fundamental multithreaded implementation is possibly only matched in importance by our display server being switched to Wayland, this allows Wayland to do OpenMP which hell yes is good!!!
                    Any project using LLVM+static analysis is an overwhelming win in all respects to show amazing code errors and obscure bugs that could have never been found otherwise.

                    In consideration this also allows LLVM pipe and its other projects to do some easy and simple multithreaded work to help do display pipeline compilation stages to be multicore far easier and so many things once leveraged.

                    I'm one happy camper it is finally complete, the faster it gets merged into mainline the better! He deserves at least a few beer for such a great task completed!

                    Comment

                    Working...
                    X