AMD releases APPML source code, creates clMath library

For the past three years, the AMD library team has been heads down, working on an important project – the Accelerated Parallel Processing Math Library.  APPML contains an OpenCL implementation of BLAS and FFT routines. The library enables developers to accelerate common scientific and engineering computations on APUs and discrete graphics accelerators. Under the right conditions, we can accelerate them A LOT.  After 6 caffeine-powered releases, I’m proud of the work we accomplished.  Along the way, we received numerous requests to participate in the library.  We heard you, and we believe that the time is right to take the project to the next level.

It brings me great pleasure to announce that in close collaboration with AccelerEyes, we are opening the source to the APPML project and making it available on GitHub under the name clMath. AccelerEyes engineers are dedicating significant resources for continued development of the clMath library, and we intend for this project to be a focal point for collaboration.  With AccelerEyes, we welcome adoption and encourage contributions to the source.  We look forward to your voice in this conversation.

The clMath source will be licensed under the Apache License, Version 2.0  The source also includes our test and performance infrastructure.  Our hope is that the community will embrace and improve the existing code, and keep the projects going for years to come!

You will find the new clMath projects at the following URL’s:

and we have also created two mailing lists for the clMath projects to help facilitate communication between users and developers of the libraries respectively:

Both projects compile for Linux and Windows and the associated project wiki pages contain build instructions and other related project documentation; interested developers should make sure to scan through the wiki.

Kent Knox has spent 15 years as a developer at AMD and has served as the technical lead for the APPML and Bolt projects. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites, and references to third party trademarks, are provided for convenience and illustrative purposes only. Unless explicitly stated, AMD is not responsible for the contents of such links, and no third party endorsement of AMD or any of its products is implied.

11 Responses

  1. Sebastian

    Kent, this is awesome. An open-source FFT/BLAS library will open the possibility of merging FFT/BLAS kernels with other kernels. We should now also be able to tweak functions for specific use cases if we so desire.

    I was wondering if there is a chance to test-drive an AMD GPU using OpenCL. So far we’re an Nvidia shop only, but if I can show that AMD GPUs are equivalent or surpass Nvidia in terms of performance (FFT+BLAS), we might just order a few AMD GPUs. You can contact me under sschaet – at – gwdg dot de.


  2. kknox

    Thank you for leaving your feedback Sebastian; I have forwarded your comment on to our business relationship guys. If you have technical questions or comments, leave feedback on our GitHub repositories or our APPML forums ( and I’ll see the comments there.


  3. Alan

    Thank you so much for releasing this (especially for releasing the source!). At a previous job, we found that a 7970 blew away the Tesla and GTX 680 cards we’d previously been using for some DX11 computer vision applications (20-40% performance improvement depending on the particular algorithm).

    In my current job, we’ve been using CUDA mainly because of the lack of an official BLAS package for OpenCL. Now that this is out, I think it’s feasible for me to do some benchmarking to see if switching to AMD cards will be a win for us. Please continue turning out useful stuff like this!