Portland Group, PGI for GPU-supported HPC systems
Accelerator Fortan and C Compiler Roadmap Announced


June 25, 2009 - - The Portland Group®, a wholly owned subsidiary of STMicroelectronics, announced the official release of the PGI® Release 9.0 line, a high-performance parallelization compiler and development tool for Linux, Mac OS X, and Windows.

PGI Release 9.0 is the first product released to support the advanced PGI Accelerator™ programming model on x64 processor-based Linux systems that use GPUs with NVIDIA CUDA support.

In high-performance computing (HPC), the trend of using GPUs (Graphics Processing Units) to enhance the performance of general-purpose accelerators is increasing. Until now, using GPUs for Fortran applications has been extremely limited.

Developers targeting GPU accelerators had to program in C at a high level of detail using sequences of function calls to manage data movement between the x64 host and the GPU and to offload computations from the host to the GPU.

The PGI Accelerator Fortran and C compilers automatically analyze the overall program structure and data, partition the application portion between the multicore x64 CPU and GPU as specified by user commands, and define and generate loop mappings to automatically utilize the parallel cores, hardware threading, and SIMD vector capabilities of advanced GPUs.

Douglas Miles, Director of the Portland Group, said, "PGI 9.0 has opened up the possibility of general-purpose programming on NVIDIA CUDA-supported GPUs for professionals in science and engineering fields." Programming models such as CUDA and OpenCL enable expert programmers to write efficient GPU code by revealing low-level details of the hardware.

"The PGI Accelerator model allows any Fortran or C programmer to incrementally add GPU optimizations programmatically using compiler instructions similar to those used for OpenMP programming for multi-core CPUs," he said.

The PGI Accelerator programming model is available to all PGI 9.0 license users for free as an extended preview in the PGI Fortran and C compilers on Linux until the end of 2009.

Additional new features in PGI 9.0 include numerous Fortran 2003 incremental functions, Intel Xeon EX (Nehalem) optimizations including SSE 4.1/4.2 support, Six-Core AMD Opteron (Istanbul) support and optimizations, comprehensive OpenMP 3.0 support for C++, a completely new graphical interface on the PGDBG OpenMP/MPI debugger, improvements to the PGPROF performance profiling environment, and support for Fedora Core 10/11, SuSE 11.1, and Ubuntu 9.


PGI Accelerator Compiler Product Line Announcement
Portland Group also announced plans for a new product line of x64+GPU compilers and tools. These compilers and tools will be available starting with products scheduled for release in the fourth quarter of 2009. The PGI Workstation Accelerator product line is expected to provide the following features.

• PGI Accelerator Fortran and C compiler for Linux, MacOS, and Windows
• Support for GPU device resident data in PGI Accelerator programming models
• x64+GPU Application Profiling using PGPROF Profiler
• Generates Extended Compiler Feedback (CCFF) to support application porting and optimization targeting x64+ GPUs