Scout: High-Performance Heterogeneous Computing Made Simple

Abstract

Researchers must often write their own simulation and analysis software. During this process they simultaneously confront both computational and scientific problems. Current strategies for aiding the generation of performance-oriented programs do not abstract the software development from the science. Furthermore, the problem is becoming increasingly complex and pressing with the continued development of many-core and heterogeneous (CPU-GPU) architectures. To acbieve high performance, scientists must expertly navigate both software and hardware. Co-design between computer scientists and research scientists can alleviate but not solve this problem. The science community requires better tools for developing, optimizing, and future-proofing codes, allowing scientists to focus on their research while still achieving high computational performance. Scout is a parallel programming language and extensible compiler framework targeting heterogeneous architectures. It provides tbe abstraction required to buffer scientists from the constantly-shifting details of hardware while still realizing higb-performance by encapsulating software and hardware optimization within a compiler framework.

The Q Continuum Simulation: Harnessing the Power of Gpu Accelerated Supercomputers

“…Our code can also be compiled to Thrust's OpenMP or Intel TBB backends to run on multi-core CPUs (including the Xeon Phi). While Thrust essentially functions as a source-to-source translator, it may be possible to provide even more efficient support for data-parallelism using, for example, compiler optimizations (Jablin et al 2011). We use Thrust because it is a readily-available, easy-to-use, production-quality, open-source, and effective library, but our data-parallel algorithms should also be compatible with alternative data-parallel frameworks.…”

The ARES High-level Intermediate Representation

“…We modified Clang's front-end including extending the lexer to add new keywords, added new AST types to represent meshes and parallel for, extended the parser and semantic analyzer to support our new statements, declarations, and expressions, and implemented their associated code generation to IR and interface to our runtime. We created an extension of DWARF to recognize Scout constructs and modified Clang and LLDB to support debugging of Scout statements and expressions [6,12]. Scout, like the current OpenMP functionality in Clang, differs from the previous approaches in that there is no new intermediate representation: the functionality is wired into the Clang front-end and generates LLVM IR and runtime library calls directly.…”