About Ruyk

Ruyman Reyes has been working in HPC since 2007, when he joined the Research Support Unit at the University of La Laguna as system administrator and user support. He currently works at Codeplay Software Ltd, leading ComputeCpp (Codeplay's SYCL implementation) and contributing to C++ and SYCL standards. He has experience in compiler and runtime technologies, GPU programming and code optimisation. In his free time he loves hiking, swimming and sightseeing. At some point in his past life he played piano and guitar, and he would like to do it again in the near future.

accULL release 0.3

Well, after a summer break and some Hangout meetings with the team, it is time again to publish a new release of the accULL. Release 0.3 (codename: Lucas) has many bugs fixed in the compiler and in the toolchain. Although we circulated an alpha a while ago, we have been adding features and fixing bugs thanks to the feedback we have received from users.

Some people reported problems when using automake 1.4 while building the runtime, so we updated the autoconf and automake files so that it is no longer using subdirs but a cleaner way to produce the Frangollo library. It is, however, a static library. It should not be difficult to build a dynamic library, and we may add that option to a future release.

We have improved the implementation of the parallels directive. In the previous release it used a default kernel launch configuration of 16 threads per grid dimension.  Now we are using the same estimator we use for the kernels region and users should see at least some performance improvements.

We have now an accull repository and a project webpage (in bitbucket of course!) were you can get the latest versions and information about the project. You can get the latest released package from the Downloads package, or if you are feeling in the mood for and adventure, you can download the development repository, which will get the latest version of Frangollo and YaCF from their respective repositories. Feel free to experiment and report issues or feature requests!

accULL release 0.2

It’s been a while since we last published an update, but it does not mean we have not been working!

Thanks to the benchmark codes provided by Nick Johnson (EPCC), we have been able to detect several situations that were not properly addressed by our implementation. We also took the opportunity to do some “house cleaning”, and we added a set of tests with an auto-verification script to help the maintenance of the code. Since the 0.1a release that we published on October 11 2012, we have committed around 50 changesets to the compiler and more than 20 to the runtime, so we believe it is that time of the year when we pack everything, write some documentation and release a new version, release 0.2. Still far from version 1.0, but getting closer.

Many thanks to all the people that has contributed to this release, in particular Juan J. Fumero and José L. Grillo, from University of La Laguna, who have been doing an incredible job to have everything ready for this release!

The new version can be downloaded here. Follows is a list of relevant issues added or fixed in this new release:

  • Added 20 new validation tests
  • Added an script to run and check the tests automatically
  • Added support for the acc parallel  directive, including num_gangs and num_workers
  • Improved support for the if clause
  • Many minor/weird bugs fixed both in compiler and runtime
  • Added suport for the firstprivate clause (including arrays)
  • Removed the requirement of putting reduction variables in a copy clause before using it
  • Script to ease the compilation of source code (just type ./accull source.c)
  • Some cleanup in the Frangollo code generator
  • Added support for Kepler cards to the runtime
  • Code generation should be slightly faster than before

As you can see, the majority of the changes have affected the compiler. We expect a new release with many changes to the runtime, addressing cleanup and performance, in a short period of time. Keep posted!

Running OpenACC without a GPU: a sneak peek of accULL

In our research group (GCAP)  we have been working on directives for GPUs for the last three years (MICPRO-2011,JoS-2011). When the OpenACC standard was announced in the SC11, we found plenty of similarities with the language we were working at the moment. Immediately we focused our work in supporting and improving the new standard.
Although the amount of work to fully implement the OpenACC standard is not negligible, our compiler framework, yacf, provided us with the required flexibility to build a parser and a lowering phase to our runtime in a couple of weeks.

Most of our recent work will appear on conferences in the upcoming months. We will be presenting contributions about accULL in the following conferences:

Feel free to speak with us there. We will provide slides and detailed information in the upcoming weeks.

Source code of the compiler framework is already available , and if you are interested we can provide the development version of accULL. A public repository will be available in the near future. On the meantime, we show here a short “teaser” of the OpenCL support that we’ve implemented in the runtime.

Picture below shows execution time for a Molecular Dynamics simulation implemented in OpenACC, running on top of one of our development servers, M4. M4 is a shared memory system (that’s it, no GPU at all) with 4 Intel(R) Xeon(R) E7 4850 CPU processors.

The usual approach to implement algorithms for these architectures is to use OpenMP, however, using the Intel OpenCL platform it is also possible to run OpenACC on top of the server.

Red bars shows the execution time of OpenMP (provided by GCC 4.4.5) and green bars shows the execution time of the same code using accULL OpenCL backend.

In this case, the runtime library detects that it is possible to avoid transfers and uses mapping instead, thus avoiding unnecessary memory transfers. The Intel OpenCL platform takes advantage of the AVX instruction set, and the kernel is nicely transformed on vector instructions, which execute natively on the CPU.

accULL can also be used to run OpenACC programs while you are traveling (which is sometimes useful!). The following figure shows the same molecular dynamics simulation running on my  laptop.

Using an environment variable, runtime can switch between using the internal GPU or the CPU (again using the Intel OpenCL platform). The largest problem size froze my laptop, the problem was too big for 512Mb of graphics memory.

I hope this gives you an idea of the kind of work we are doing within accULL. More information will be available on the following weeks!