Michael Wolfe's Programming Guide
The PGI Accelerator Programming Model on NVIDIA GPUs—Part 2 Performance Tuning
Part 1 of this series introduced the PGI Accelerator Programming Model, showed three simple programs in C and in Fortran, and presented a few details for building and running a program on the GPU.
Part 2 looks at issues affecting performance including how to recognize and address them. It includes an in-depth look at the four most important performance issues: writing an appropriately parallel algorithm, tuning the data movement between the host and the accelerator, tuning memory loads and stores on the accelerator, and tuning the loop schedule.
Also in this issue…
Tool Tips: A New Direction for PGI Performance Profiling
This article introduces the new easy-to-use performance data collection tool PGCOLLECT.
Cross Platform: Using Microsoft MPI with PGI Visual Fortran
Release 9.0 of PGI Visual Fortran (PVF) and PGI Workstation support building, running and debugging MPI applications on Windows laptops and clusters.
Tutorial: Building Cactus BenchADM with PGI Accelerator Compilers
A step by step guide to using the PGI Accelerator Fortran compiler to build a GPU enabled version of BenchADM, the computational kernel representative of many applications in numerical relativity.
Case Study: Speed-Up for Physicists
Dr. Stephan Raub at Dusseldorf University describes his experience in using PGI Accelerator compilers to port an Iterative Summation of real-time Path Integrals algorithm to NVIDIA GPUs.
Upcoming Events
A one day tutorial session GPU Programming with the PGI Accelerator Programming Model and with PGI CUDA Fortran. Monday, 16 November in Portland. Cost is $500 per person and includes a copy of PGI Accelerator Fortran/C/C++ Workstation compilers & tools.
HPC on Wall Street 14 September, Roosevelt Hotel, NYC.
New Resources
PGI x64+GPU Fortran and C99 Compilers IEEE webinar
PGI x64+GPU Fortran and C99 Compilers IEEE webinar slides
PGI Fortran & C Accelerator Programming Model white paper
CUDA Fortran Programming Guide and Reference
Recent News
PGI Visual Fortran 9.0 adds MSMPI debugging support
New PGI 9.0 Compilers Simplify x64+GPU Programming.
What's new in PGI Release 9.0.
PGI and NVIDIA Team To Deliver CUDA Fortran Compiler
Next Issue Highlights
PGI Accelerator Programming Model—Part 3: Porting a Large Application
CUDA Fortran Revealed
Profiling non-PGI built executables with PGPROF
PGI Accelerator Road Map
The PGI multi-platform network floating license enables you to run PGI compilers and tools on any combination of Linux, Mac OS X and Windows systems simultaneously up to the maximum number of seats authorized by your license. Contact PGI Sales for details on upgrading.
- The Portland Group, Inc.
- Suite 320
- Two Centerpointe Drive
- Lake Oswego, OR 97035