Announcement

Collapse
No announcement yet.

ArrayFire Accelerated Compute Library Open-Sourced

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

  • ArrayFire Accelerated Compute Library Open-Sourced

    Phoronix: ArrayFire Accelerated Compute Library Open-Sourced

    The ArrayFire GPU compute library that allows for simplified GPU computing via targeting its own optimized library and API for GPGPU kernel generation than writing your own CUDA/OpenCL kernels, has been open-sourced...

    http://www.phoronix.com/vr.php?view=MTg0Njk

  • #2
    Hello from ArrayFire

    First of all, thank you for writing about ArrayFire going open source. I am the lead engineer of the library. There are a few things to clarify.

    - The benchmarks showing us faster than MKL / Armadillo / Eigen were when we were a GPU only library. The C backend should be on par with the aforementioned libraries.
    - It would be great if you can link to our github page: https://github.com/arrayfire/arrayfire
    - The open source project required refactoring our old codebase to make it more modular. There are some kinks we are ironing out. Please follow us on github or twitter (@arrayfire) for more news.

    Comment


    • #3
      AFAIK you need to buy Pro version to have 'double precision', so this is mostly being convenient only as multi device kind of wrapper. With given feature set I would still pick handwriting or maybe CUSP/Eigen.

      Comment


      • #4
        Originally posted by magika View Post
        AFAIK you need to buy Pro version to have 'double precision', so this is mostly being convenient only as multi device kind of wrapper. With given feature set I would still pick handwriting or maybe CUSP/Eigen.
        The Pro version was required only for double precision linear algebra functions. This was mostly because we were using a non-free linear algebra library. All of ArrayFire is now open source, but we currently cannot offer those linear algebra functions. We are currently trying to find alternatives for the missing LA functionality.

        Comment


        • #5
          Originally posted by umar456 View Post
          The Pro version was required only for double precision linear algebra functions. This was mostly because we were using a non-free linear algebra library. All of ArrayFire is now open source, but we currently cannot offer those linear algebra functions. We are currently trying to find alternatives for the missing LA functionality.
          Hi, is there any fortran support?
          What are recommended compilers?
          Also is it possible to mix it with MPI?
          Does it support sparse matrices?

          Anyway thanks for open soursing it.
          Rakot
          Phoronix Member
          Last edited by Rakot; 24 November 2014, 10:10 PM.

          Comment


          • #6
            Originally posted by Rakot View Post
            Hi, is there any fortran support?
            What are recommended compilers?
            Also is it possible to mix it with MPI?
            Does it support sparse matrices?

            Anyway thanks for open soursing it.
            We will have support for fortran through wrappers. You can find wrappers for many languages at our GitHub page. We will be updating them for the new C interface which will provide additional functionality (JIT, etc.) in the near future.

            You will need at least GCC >= 4.7, Clang, and Visual Studio 2013 to build ArrayFire. You should be able to use the binaries with older compilers.

            ArrayFire gives you direct access to the device and system pointers through our interface. You should be able to use MPI functions to communicate the data to other nodes.

            We do not currently have support for sparse matrices.

            We are always interested in hearing from our community. You can find us on GitHub and at ArrayFire.com. We look forward to hearing from you.

            Comment


            • #7
              If there gets to be a python release of this I will be very intrigued. I hope to see this gain popularity, due to its portability and ease of use.

              Comment


              • #8
                Originally posted by schmidtbag View Post
                If there gets to be a python release of this I will be very intrigued. I hope to see this gain popularity, due to its portability and ease of use.
                Someone actually created a Python wrapper for ArrayFire here

                Comment


                • #9
                  Originally posted by Rakot View Post
                  Hi, is there any fortran support?
                  What are recommended compilers?
                  Also is it possible to mix it with MPI?
                  Does it support sparse matrices?

                  Anyway thanks for open soursing it.
                  Yes we do have fortran support. https://github.com/arrayfire/arrayfire_fortran

                  All you need is arrayfire and gfortran.

                  However there have been a few API changes since we went open source. We are going to sanitize things in the near future.

                  Comment


                  • #10
                    Originally posted by umar456 View Post
                    The Pro version was required only for double precision linear algebra functions. This was mostly because we were using a non-free linear algebra library. All of ArrayFire is now open source, but we currently cannot offer those linear algebra functions. We are currently trying to find alternatives for the missing LA functionality.
                    Would a BLAS or LAPACK library (like OpenBLAS , MPACK , ATLAS, or CBLAS) be what you're looking for, or do you need something else?
                    http://en.wikipedia.org/wiki/Basic_L...mplementations might be helpful, if that's the sort of thing you want.

                    Comment

                    Working...
                    X