Their latest post reads, "Today the Radeon Technology Group is releasing a preview version of the Radeon Open Compute Kernel driver (ROCK) and Radeon Open Compute runtime ROCR, allowing the exploration of what is possible with the open GPU computing foundation. The objective of this release is to start a dialog with the commercial and academic HPC communities that will shape the future direction of the Boltzmann Initiative, both for the coming year and beyond. We are excited to present to you our first public release of the Boltzmann driver and runtime with HCC and HIP."
This initial support features the HCC compiler with C/C++ support, the HIP runtime, Mesa X11 and OpenGL support, multi-GPU support, and more.
The initial release only works with the AMD R9 Fury "Fiji" graphics processors. Additionally, they recommendusing a "Intel Haswell or newer" processor whether it be a Core i5/i7 or Xeon CPU with no apparent support for their own AMD processors.
Features coming in the future include peer-ot-peer multi-GPU support, HCC built on the Lightning Compiler with native GCN ISA code generation, the GCN ISA assembler/disassembler, process concurrency and preemption, and system manageability tools.
While the initial release of this Radeon Open Compute ROCK/ROCR platform was announced, I don't see any link from their announcement nor have I been able to find it yet in any GitHub repository.
Update: GPUOpen's GitHub account is now RadeonOpenCompute where they have published this initial code.