Portland Group has released v9.0 of its PGI line of parallelization compilers and development tools. The update is the first release to include support for the PGI Accelerator programming model on x64-based Linux systems utilizing NVIDIA CUDA-enabled GPUs. The PGI Accelerator Fortran and C compilers are designed to automatically analyze a program's structure and data as well as to split portions of the application between a multi-core x64 CPU and a GPU. The portions split between each device are controlled by a user's preferences. The PGI Accelerator additionally defines and generates maps of loops, enabling parallel cores to be used simultaneously, with support for hardware threading and SIMD vectors.

The inclusion of CUDA and Open CL programming models allows programmers to write more efficient GPU code by utilizing low-level details of the hardware. The accelerator also enables Fortran and C programmers to add GPU optimizations to a program using compiler directives similar to Open MP programming for CPUs. PGI 9.0 also features Intel Xeon EX optimizations, along with support for SSE 4.1 and 4.2, Six-Core AMD Opteron, Fedora Core 10 and 11, and OpenMP 3.0 in C++.

Evaluation copies of PGI 9.0 Fortran and C compilers are being previewed for free for the remainder of 2009 from the Portland Group website.