SALT LAKE CITY, Nov. 15 – The Portland Group (PGI), a wholly-owned subsidiary of STMicroelectronics and the leading independent supplier of compilers and tools for high-performance computing, today announced plans to extend its PGI Accelerator compiler technology with OpenACC to Intel Xeon Phi coprocessors, Intel's family of products based on the Intel Many Integrated Core (MIC) architecture. Currently, scientists and engineers are using PGI Accelerator Fortran and C compilers as a means to exploit the massively parallel throughput capacity of CUDA-enabled GPUs from NVIDIA. Using their existing code bases and with only minor modifications to their build scripts, software developers using PGI Accelerator compilers will soon be able to target Intel Xeon Phi coprocessors as well.

"Until now, industry CIOs and government and university lab directors have encountered difficult decisions when evaluating accelerator technologies. Do we wait to see which technologies win out? Which programming model or models do we choose? How do we retrain developers to use these programming models? Will our applications be future-proof and portable across the different types of host CPUs and coprocessors?" said Douglas Miles, director, The Portland Group. "PGI Accelerator compilers will make Xeon Phi coprocessors programmable using standard OpenACC directives that are fully compatible with the accelerator application development efforts already under way at most significant HPC centers and sites."

First announced in mid-2009, PGI Accelerator compilers provide a high-level coprocessor programming model intended for scientists, engineers and other domain experts who aren't full-time programmers. PGI Accelerator Fortran 2003, C and C++ compilers enable programmers to offload compute-intensive portions of an application to a coprocessor by adding portable compiler directives, treated as comments by other compilers, to existing standard-compliant programs and recompiling with appropriate compiler options. In 2012, the PGI Accelerator compilers were enhanced to support the new de facto standard OpenACC directives.

Key advantages to programming coprocessors with OpenACC directives include:

• >High-level ? No prior experience with OpenCL, CUDA or other low-level programming models required. Ideal for domain experts.
• Efficient ? Well-designed algorithms that would perform well when written using a low-level programming model can perform just as well using directives.
• Single source base ? No separate co-processor source code required. Compile the same program for multi-core CPUs using PGI or any other standard-compliant compiler.
• Performance portable ? Supports GPU accelerators and co-processors from multiple vendors.
• Incremental ? Developers can port and tune parts of their application as time permits. No wholesale rewrite required.
• Quick ? Most developers see results with modest effort.

"PGI OpenACC will enable programmers to develop portable applications that maximize the performance and power efficiency benefits of the hybrid CPU/Accelerator architecture of Titan," said Buddy Bland, Titan Project Director, Oak Ridge National Laboratory. "We are pleased to see that PGI is adding support for Intel Xeon Phi, which ensures that accelerated applications developed for Titan will port readily to HPC systems that incorporate Intel co-processors."

"As part of our on-going assessments of future computing architectures Sandia is actively engaged in the evaluation of solutions from a spectrum of vendors including Intel, NVIDIA, AMD, ARM and IBM," said Simon Hammond, Sandia National Laboratories. "We are already using PGI's compilers for exploratory code development on accelerators from NVIDIA. PGI's extension of OpenACC to support Intel's Xeon Phi processors enables us to use an open standards based programming model across a wide range of accelerator hardware."

"I've been working with PGI off-and-on since 2008 to explore programming models that can address both today's multi-core and tomorrows accelerator-enabled compute nodes," said John Michalakes, HPC Scientist, National Renewable Energy Laboratory. "PGI's commitment to creating programming models and compilers that enable functional and performance portability across platforms is unquestioned, and I look forward to using their compilers on Xeon Phi."

"We have been studying the performance of the FIM and NIM Numerical Weather Prediction models on both NVIDIA GPUs and Intel Xeon Phi co-processors," said Tom Henderson, Global Systems Division, at NOAA's Earth System Research Laboratory in Boulder. "We have already used the PGI compilers for our NVIDIA work and are pleased to see PGI will be supporting OpenACC as a uniform and standard programming model across both types of accelerators. We are hopeful that PGI OpenACC will provide a performance-portable solution that allows maintenance of a single source code for GPU, CPU, and Xeon Phi."

Beta support for the OpenACC standard on Xeon Phi coprocessors is planned for a PGI release in the first half of 2013. It will be available free of charge to PGI Accelerator licensees with a current PGI subscription. PGI Accelerator Fortran, C and C++ compilers with OpenACC directives currently support x64+NVIDIA systems running under Linux, OS X and Windows; the compilers are supported on all Intel and AMD x64 processor-based systems with CUDA-enabled NVIDIA GPUs. More information on the PGI Accelerator compilers with OpenACC is available at www.pgroup.com/accelerate. More information on the OpenACC API and standard can be found at www.openacc.org. More information on the Intel Xeon Phi coprocessor is at www.intel.com.

-----

Source: The Portland Group