Technical Information from The Portland Group
February 2010
Michael Wolfe

Michael Wolfe's Programming Guide

The PGI Accelerator Programming Model on NVIDIA GPUs—Part 4: New Features

Part 4 in the series looks at the newest PGI Accelerator compiler features in PGI 2010. In response to user requests, this latest release includes support for leaving data on the GPU between kernels by using data regions and support for GPU resident reductions. Other new features covered include performance profiling, new API routines and a number of other enhancements. The article concludes by looking at some of the new features coming in updates over the next few months including expanded support for the PGI Unified Binary™ and support for NVIDIA Fermi-class GPUs.

Also in this issue…

Using PGPROF and the CUDA Visual Profiler to
Profile GPU Applications

Tools and techniques for profiling both PGI Accelerator and CUDA Fortran programs.

CUDA Fortran Data Management

An overview and example of managing both CPU and GPU data using CUDA Fortran.

Tuning a Monte Carlo Algorithm on GPUs

A step-by-step example demonstrating many useful CUDA Fortran techniques including device resident data, sum reductions, data transfer optimization and calling a CUDA C kernel from CUDA Fortran.

Understanding the CUDA Data Parallel Threading
Model—A Primer

Early approaches to computing on GPUs cast computations into a graphics framework. With the release of NVIDIA's CUDA, data parallel computing on a GPU became more straightforward. Here, we'll describe the data parallelism model supported in CUDA on NVIDIA GPUs.

Upcoming Events

A one day tutorial GPU Programming with CUDA Fortran and C and the PGI Accelerator Programming Model. Sunday, 30 May in Hamburg. Cost includes a discounted copy of PGI Accelerator Fortran/C/C++ Workstation compilers & tools.

Register now for the free PGI CUDA Fortran Introduction webinar March 24. Sponsored by NVIDIA.

Resources

PGInsider back issues

PGI Fortran & C Accelerator Programming Model white paper
(v1.1, Nov. 2009)

CUDA Fortran Programming Reference
(v1.0, Nov. 2009)

PGI Accelerator Licensing FAQ

PGI Accelerator User Forum

Recent News

PGI CUDA Fortran Released

PGI 2010 Released

Next Issue Highlights

PGI Accelerator Programming Model—Part 5: CUDA 3.0 and Fermi

Writing CUDA Fortran Kernels

GPU Programming from Within Visual Studio Using PGI Visual Fortran

A Survey of Codes Ported to GPUs with PGI Compilers and Tools

Did you know…IMSL

In cooperation with Rogue Wave and Visual Numerics, Inc., a Windows version of the IMSL Fortran library precompiled with PGI Fortran is available directly from PGI. A precompiled IMSL Fortran library for PGI Fortran on Linux is available from VNI. See the IMSL pricing page on the PGI website or contact PGI Sales for a quote or for additional information.

  • The Portland Group, Inc.
  • Suite 320
  • Two Centerpointe Drive
  • Lake Oswego, OR 97035