Conference:Multicore Challenge 2014 (click here to see full programme)
Speaker:David Stuttard, Software Manager – Compilers
Presentation Title:“CM” a domain specific language for the Intel GPU architecture
Abstract:Intel’s GPU architecture, whilst supporting standard languages targeting multicore systems, also provides powerful support for features not natively supported in these languages. “CM” is a lower level proprietary DSL that supports a different model for parallelism that provides direct support for hardware features particularly suited to media and compute tasks. In particular it allows lower level and more flexible access to the hardware’s parallelism allowing different parallelisation within the same algorithm as often required by for instance modern video codecs.
This presentation describes some of these hardware features, how the CM language supports them, how the programmer can make use of these features and how it sits within a wider Media Developer Framework runtime environment for multicore programming.Topics covered in the presentation will include:

  • CM Domain Specific Language
  • Media and compute acceleration
  • Intel GPU Architecture
BiographyDavid has over 20 years of experience developing multicore tools and architectures. David was one of the original architects for the PixelFusion massively parallel GPU as well as developing simulation and programming tools for this architecture. Subsequently David was one of the co-developers of the ClearSpeed multicore accelerator processors for which he designed a programming language and associated compiler tools.David also developed LLVM based OpenCL and OpenGLES compiler tools for the ZiiLABS range of ARM based processors for embedded systems (targeting the parallel GPU processor designed by ZiiLABS integrated into the SoC). David is currently at Intel working on parallel languages and compilers for GPU architectures

Slides and recording unavailable at the request of the speaker