September 2010
Michael Wolfe's Programming Guide
CUDA Fortran: The Next Level
This article describes the new kernel directive coming to CUDA Fortran with PGI Release 2011 due out in November. The kernel directive will make porting to the CUDA programming model more straightforward and let developers concentrate their efforts on the time-critical portions of their application.
Also in this issue…
Porting the SPEC Benchmark BWAVES to GPUs with CUDA Fortran
The paper covers many important aspects of porting a typical HPC application to CUDA including identifying CPU bottlenecks, distributing execution to both the CPU and GPU, using asynchronous data transfers to minimize overhead and writing efficient CUDA kernels. The paper concludes with accuracy and performance results across several platforms.
Accelerating Visual C++ Applications on GPUs
Until now, a directive-based GPU programming model has not been accessible to Windows Visual C++ developers. The PGI Accelerator C plug-in can be used to target specific files for GPU compilation from within Microsoft Visual Studio.
Using the CULA GPU-enabled LAPACK Library
with CUDA Fortran
This article looks at a new version of the CULA GPU-accelerated linear algebra library specifically for CUDA Fortran. Includes information on how to use the library, performance data and a link to a free CULA download.
Upcoming Events
Concurrent with SC10 in New Orleans, Michael Wolfe will be conducting a one day tutorial on "GPU Programming with CUDA Fortran, CUDA C and the PGI Accelerator Programming Model" Monday, 15 November 2010. The cost is $500US. Attendees may also purchase a license to the PGI Accelerator Workstation product of their choice for just $250 additional.
Resources
PGI Accelerator C99 & Fortran programming model specification
CUDA Fortran Programming Reference
Recent News
The Portland Group releases PGI Visual Fortran for Visual Studio 2010
US academics qualify for a free license to PGI Visual Fortran
Next Issue Highlights
PGI Unified Binary for Intel and AMD CPUs and NVIDIA GPUs
Performance Tuning Using PGPROF and Hardware Counters
Scaling CUDA Fortran Applications for Multiple GPUs
Object Oriented Programming with Fortran 2003
CULA is a cross-platform, cross-language GPU-accelerated linear algebra library. EM Photonics, has just released a version of CULA with an interface to CUDA Fortran.
- The Portland Group, Inc.
- Suite 320
- Two Centerpointe Drive
- Lake Oswego, OR 97035