Offload – Automating Code Migration to Heterogeneous Multicore Systems

Abstract

Abstract. We present Offload, a programming model for offloading parts of a C++ application to run on accelerator cores in a heterogeneous multicore system. Code to be offloaded is enclosed in an offload scope; all functions called indirectly from an offload scope are compiled for the accelerator cores. Data defined inside/outside an offload scope resides in accelerator/host memory respectively, and code to move data between memory spaces is generated automatically by the compiler. This is achieved by distinguishing between host and accelerator pointers at the type level, and compiling multiple versions of functions based on pointer parameter configurations using automatic call-graph duplication. We discuss solutions to several challenging issues related to call-graph duplication, and present an implementation of Offload for the Cell BE processor, evaluated using a number of benchmarks.

Accelerating Code on Multi-cores with FastFlow

“…In the case of GPGPUs those (low-level) libraries include Brook [13], NVidia CUDA, and OpenCL. At a higher-level, Offload [14] enables offloading of parts of a C++ application, which are wrapped in offload blocks, onto hardware accelerators for asynchronous execution; OMPSs [15] enables the offloading of OpenCL and CUDA kernels as an OpenMP extension [16]. FastFlow, in contrast with these frameworks, does not target specific (hardware) accelerators but realizes a virtual accelerator running on the main CPUs and thus does not require the development of specific code.…”

Self-Adaptive OmpSs Tasks in Heterogeneous Environments

“…These codelets can either be hand-written for a specific architecture or be generated by some code generator. Offload [18] is a programming model for offloading portions of C++ applications to run on accelerators. Code to be offloaded is wrapped in an offload block, indicating that the code should be compiled for an accelerator, and executed asynchronously as a separate thread.…”

Mainstream Parallel Array Programming on Cell

“…E is a source to source compiler, translating from the 'F' subset of Fortran 95 to Offload C++ [6]: a C++ language extension utilising pointer locality. The compiler targets heterogeneous multicore architectures, and in particular the CBE.…”

“…E translates from 'F' to Offload C++, a C++ language extension and runtime library [6] targeting heterogeneous architectures. The most prominent language feature of Offload C++ is the offload block which provides a traditional 'C' compound statement, prefixed with the keyword offload, to be executed asynchronously to the main thread.…”