Hi!
There was a session by Greg Stoner of AMD, and in the news report by Phoronix it can be read that soon there will be a release of LLVM >>HSAIL backend, as well as a C++AMP compiler front-end to Clang. Does this mean, that perhaps the only thing missing (shortly) for C++AMP to run portably under linux on the desktop is an LLVM-SPIR compiler back-end and that vendors implement their SPIR compiler inside their runtime?
It has been said in the OpenCL part of this forum that SPIR compiler support is a top priority for SDK 2.9, which could serve as the first basis of testing of the LLVM-SPIR back-end. I know that this forum is no place to give information on ongoing projects, but it would be nice if AMD could also give boost to the LLVM-SPIR backend. C++AMP is certainly one of the most promising attempts at leveraging GPGPU power. Having read through the slides HSA is certainly more flexible and powerful, completely blurring the boundary between the CPU and GPU. HSA if I understood correctly is virtually capable of enabling apps (even a complete OS) to be able to run dominantly on a GPU, with the CPU as a co-processor. Naturally an OS is mostly latency critical, so this would be useless, but having the GPU being able to reissue work for itself and launch CPU functions, practically the asymmetry of the CPU-GPU could even be reversed in some applications. C++AMP could be an elegant way to make use of these features.
Are these assumptions correct, or did I fatally miss the point? As far as I understood HSA enables the IGP to act as a full fledged compute unit, providing the same level of security and coherence as the CPU alone.