A Taxonomy of Modern GPGPU Programming Methods: On the Benefits of a Unified Specification

Abstract

Several Application Programming Interfaces (APIs) and frameworks have been proposed to simplify the development of General-Purpose GPU (GPGPU) applications. GPGPU application development typically involves specific customization for the target operating systems and hardware devices. The effort to port applications from one API to the other (or to develop multi-target applications) is complicated by the availability of a plethora of specifications, which in essence offers very similar underlying functionality. In this work we provide an in-depth study of six state-of-the-art GPGPU APIs. From these we derive a taxonomy of the common semantics and propose a unified specification. We describe a methodology to translate this unified specification into different target APIs. This simplifies crossplatform application development and provides a clean framework for benchmarking. Our proposed unified specification is called GUST (GPGPU Unified Specification and Translation) and it captures common functionality found in compute-only APIs (e.g., CUDA and OpenCL), in the compute pipeline of traditional graphic-oriented APIs (e.g., OpenGL and Direct3D11) and in last-generation bare-metal APIs (e.g., Vulkan and Direct3D12). The proposed translation methodology solves differences between specific APIs in a transparent manner, without hiding available tuning knobs for compute kernel optimizations and fostering best programming practices in a simple manner.

Brief Announcement: Optimized GPU-accelerated Feature Extraction for ORB-SLAM Systems

“…CUDA enables to dispatch kernels through a launch configuration, i.e. a grid specified by the programmer in which parallel GPU threads are logically organized [4] in blocks of threads. Moreover, it allows the programmer to express an added layer of parallelism through CUDA Streams and Events.…”

Optimized Local Path Planner Implementation for GPU-Accelerated Embedded Systems

“…Differently, we also ported both obstacle avoidance and the best path selection on the GPU. We use NVIDIA CUDA to exploit GPU parallelism in different ways: using parallel threads [22], exploiting Shared Memory and syncthread directive; overlapping computing and memory transfers using Streams and Events [23].…”