Future of programming accelerators is Open, wide open

The hardware changes on the horizon of 2016 may already be fairly well known but what about the shifting sands of software?
The rumour from SuperComputing is of Intel quietly sounding out the customer view of pulling OpenCL support from its Xeon Phi cards.

Why Intel would want to do this? Without insider knowledge we cannot say for sure. The answer probably lies in either a CBA regarding continued support for OpenCL on XPhi or in expected popularity growth of alternatives.

It’s clear Intel are not dropping support for OpenCL - sources have confirmed the language will continue to be used on their GPUs (graphics cards). Maybe the investment of ensuring OpenCL works well as XPhi evolves has been deemed to be high? But that would be implying that Intel do not see OpenCL dominating the future programming of high end compute.

Considering the alternatives for programming NVIDIA GPUs, we would come up with CUDA and OpenACC. We know that OpenMP will be absorbing directives from the OpenACC standard and that NVIDIA’s support of OpenCL for their compute GPUs is far from ideal.

Pulling these threads together, we would be putting our money on the ascension of OpenMP and seriously consider how safe our OpenCL investments are.