1 of 1

CCAMP: Integrated Translation and Optimization for OpenACC and OpenMP

Bi-directional translation and optimizations between OpenMP and OpenACC using CCAMP

OpenARC System Architecture

Jacob Lambert, Mohammad Alaul Haque Monil, Seyong Lee, Allen Malony, and Jeffrey S. Vetter, Leveraging Compiler-Based Translation to Evaluate a Diversity of Exascale Platforms, P3HPC: Performance, Portability & Productivity in HPC, in conjunction to SC22, 2022.

Scientific Achievement

Developed an integrated OpenACC and OpenMP interoperable framework, which allows 1) translation and optimization between OpenACC and OpenMP and 2) intermixing of OpenACC and OpenMP, to increase the portability of applications across diverse heterogeneous architectures.

Significance and Impact

High-level programming strategies such as OpenMP and OpenACC have emerged to address heterogeneous computing issues. However, the varying levels of support for these standards within vendor-specific and open-source tools, as well as the lack of performance portability across devices, have prevented the standards from achieving their goals.

Technical Approach

Develop CCAMP, which provides interoperability between OpenMP and OpenACC by facilitating language translation and directive optimization.

Language translation between the two standards
Device-specific directive optimization within each standard

Extend CCAMP to support low-level programming models such as CUDA, HIP, and OpenCL through direct translation from OpenACC/OpenMP.
Study various optimizations to better map computations to the target device in a way preferred by the back-end compilers on the given device.

TALKING POINTS:

Directive-based high-level programming models such as OpenACC and OpenMP have emerged to address the heterogeneous computing issues.
However, these existing directive standards still suffer from the performance portability issues, since different implementations have varying levels of supports and different maturity across different types of devices.
To address these issues, we developed CCAMP, a bi-directional translation and optimization framework between OpenMP and OpenACC, which provides a way for a program written in one directive model can be ported and optimized on the system that supports the other directive model, maximizing the performance portability of the existing OpenACC or OpenMP programs across heterogeneous devices.
It also allows the existing OpenACC/OpenMP programs to be automatically ported to device-specific programming models (e.g., CUDA, HIP, OpenCL) so that the program can exploit the best available device-specific programming system on the target system.

METADATA:

Project: The RAPIDS2 Institute, Scientific Discovery through Advanced Computing (SciDAC) Program.

PI name: Robert B. Ross (ANL)
Name of the program manager: ASCR research

Project: The Exascale Computing Project PROTEAS-TUNE

PI name: Jeffrey S. Vetter (ORNL)
Name of the program manager: ECP

CITATIONS:

Jacob Lambert, Seyong Lee, Jeffrey S. Vetter, and Allen D. Malony, CCAMP: An Integrated Translation and Optimization Framework for OpenACC and OpenMP, SC 2020: The International Conference for High Performance Computing, Networking, Storage, and Analysis, 2020.

DOI: 10.1109/SC41405.2020.00102

Jacob Lambert, Mohammad Alaul Haque Monil, Seyong Lee, Allen Malony, and Jeffrey S. Vetter, Leveraging Compiler-Based Translation to Evaluate a Diversity of Exascale Platforms, P3HPC: Performance, Portability & Productivity in HPC, in conjunction to SC22, 2022.

DOI: 10.1109/P3HPC56579.2022.00007