Simplifying low-level GPU programming with GAS

Abstract

Many low-level optimizations for NVIDIA GPU can only be implemented in native hardware assembly (SASS). However, programming in SASS is unproductive and not portable. To simplify low-level GPU programming, we present GAS (Gpu ASsembly), a PTX-like language that provides a stable instruction set across hardware architectures while giving programmers a low-level control of code execution. We demonstrate that GAS can be used with ease for low-level benchmarking and performance tuning in the context of Tensor Core HGEMM.