May 2010

Michael Wolfe's Programming Guide
PGI Accelerator Programming Model Support on NVIDIA Fermi GPUs
This article looks at the issues involved with developing programs using the directive-based PGI Accelerator C99 & Fortran compilers on the latest generation Fermi class GPU accelerators from NVIDIA. Topics covered include considerations for retaining portability as well as potential run-time issues. The article also looks at some of the latest compiler features recently added to support Fermi GPUs.
Also in this issue…
CUDA Fortran Device Kernels
Programming considerations when targeting NVIDIA GPUs using either CUDA Fortran or CUDA C, including how using shared memory and problem decomposition can improve performance
Using NVIDIA GPU Accelerators with
PGI Visual Fortran
An overview on accessing the PGI Accelerator and CUDA Fortran features of PGI Visual Fortran from within Microsoft Visual Studio.
Parallel Random Number Generation Using OpenMP, OpenCL and PGI Accelerator Directives
A comprehensive look at the requirements for optimizing code using four of common parallel programming technologies including OpenMP on multi-core x86 hosts and three GPU programming models: PGI Accelerator directives, CUDA and OpenCL.
OpenMP on Accelerators—A Position Paper
In the two years since PGI announced the PGI Accelerator programming model, many have suggested extending OpenMP directives to support accelerators instead of using new directives as PGI has chosen to do. This article attempts to summarize PGI's rationale.
Upcoming Events
A one day tutorial GPU Programming with CUDA Fortran and C and the PGI Accelerator Programming Model. Sunday, 30 May in Hamburg. Attendees can purchase a discounted copy of PGI Accelerator Fortran/C/C++ Workstation compilers & tools. Michael Wolfe will conduct the same one day tutorial in Paris, France on Thursday, 3 June 2010.
PGI will be presenting a half day tutorial at the DoD HPCMP 2010 Users Group Conference in Schaumburg, Illinois on Monday 14 June.
Resources
PGI Accelerator C99 & Fortran programming model specification
CUDA Fortran Programming Reference
Recent News
PGI announces support for NVIDIA Fermi architecture.
PGI compilers now available on Cray CX line of supercomputers.
Gaussian 09 for 64-bit MacOS Built with PGI.
Next Issue Highlights
Object-oriented Programming with CUDA Fortran 2003
Performance Tuning Using PGPROF and Hardware Counters
Accelerating SPEC 2006 Benchmarks using GPUs
Taking PGI CUDA Fortran to a Higher Level

If you are developing for GPUs, the PGI Accelerator Files is a new resource on the PGI website where you can find articles, tutorials, source code, benchmark data and other useful information to help you. If you have already ported your application, library or other software to GPUs, help others in the community and share your experiences.
- The Portland Group, Inc.
- Suite 320
- Two Centerpointe Drive
- Lake Oswego, OR 97035