I don't know the name of the GPUCS but the code could be added to the i915/i915kms driver for Unix
I don't know much about GPU programming, would love to learn, but I can imagine a 64-bit float is only as good as its register. But, if you combine many 64-bit floats in a line through *malloc() and using a custom size (or a multiplier in my case) could create a software 128-bit emulator, or even a 1gbit emulator. but it would be drastically slow. This is mainly for video games, but it can be applied to most types of computer programming paradigms.
Basically, I want intel to support 1024bit floats/longs/pointers through software and hardware on their GPUs and maybe cpu platforms, through an emulation layer. could save some skin in the future.