During the two-day, GTC Asia event in Beijing, NVIDIA announced plans to open source its CUDA Low Level Virtual Machine (LLVM) compiler, so to speak.
The firm has worked hard on its upcoming CUDA 4.1 release, with the software now able to utilise the LLVM to generate intermediary code that can be compiled for any platform supported by the LLVM; this coupled with the availability of the source code, could potentially see the applications of CUDA expand to new platforms beyond those developed or supported by NVIDIA or see new programming languages become available for compiling for use on NVIDIA GPUs. LLVM has been praised for its generation of high-performance intermediary code and its modular design; the Mac and iOS developers out there will have noticed that Apple has worked to adopt this compiler as the new default option when building apps for these very same reasons.
It's worth noting that NVIDIA is not open sourcing its fork of the LLVM compiler in the same way Google has Android, for example, instead developers must apply for access to the code, allowing NVIDIA to ensure firms such as AMD don't suddenly release a CUDA compatible product and that the inner workings of its GPUs are not laid bare for the world to see.
We could be seeing some interesting possibilities opening up in the future, perhaps BASIC code for running on an NVIDIA GPU? Or just perhaps, CUDA programs able to run on ARM's upcoming Mali-T6xx series of GPGPU-focused cores.
