White Paper | GPU computation in the Fabric Engine Core on HSA hardware
July 31, 2012 by Tony DeYoung
The HSA technology platform provides a heterogeneous computation platform in which both CPU and GPU cores access and manipulate memory identically. HSA enables complex data structures with pointer indirection to be shared between the CPU and GPU. Not only is unnecessary to copy of data between different memory spaces, the pointers embedded in a complex data structure are usable without change on both CPU and GPU cores.
In collaboration with AMD, the Fabric Engine development team extended the KL compiler and Fabric Engine Core execution environment to support GPU computation on high-end AMD GPUs using the LLVM back end for AMD GPU hardware. Implementation of generic GPU computation is a primary feature of the Fabric Engine Core. This will the user-created functions that are responsible for high-performance, native, parallel computation, to run, unmodified, on GPU cores.
The white paper describes a technology preview of GPU computation for Fabric Engine and its integration with the AMD LLVM OpenCL backend..