Compound Memory Models

Today's mobile, desktop, and server processors are heterogeneous, consisting not only of CPUs but also GPUs other accelerators. Such heterogeneous starting to expose a shared memory interface across these devices.Given that each individual devices typically supports distinct instruction set architecture consistency model, it is clear what the model machine should be. In this paper, we answer question by formalizing "compound" models: present compositional operational describing resulting when with models fused together. We instantiate our compound x86TSO/PTX -- CPU enforcing x86TSO GPU PTX model. A key result retains compiler mappings from language-based (scoped) C This means threads mapped device can continue use already proven C-to-x86TSO mapping, same for PTX.